A method and apparatus for identifying a type of a multimedia content publisher

Abstract

The application discloses a method for identifying the type of a multimedia content publisher, comprising: obtaining information of a multimedia content publisher to be identified; and determining the type of the multimedia content publisher to be identified based on the information of the multimedia content publisher to be identified and a predetermined target correlation relationship, wherein the target correlation relationship is used to indicate the correlation relationship between objects of multiple types, and the objects of multiple types comprise the multimedia content publisher and a target object associated with the multimedia content publisher. In the scheme, when the type of the multimedia content publisher to be identified is determined, the correlation relationship between objects of multiple types is considered in addition to the information of the multimedia content publisher to be identified, for example, the correlation relationship between multimedia content publishers and the correlation relationship between a multimedia content publisher and a target object associated with the multimedia content publisher. Therefore, the type of the multimedia content publisher to be identified can be accurately determined.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a method and device for identifying type of multimedia content publisher. BACKGROUND

[0002] In some scenarios, there is a need to identify the type of multimedia publisher. For example, when pushing multimedia content to a user, the type of publisher of the multimedia content is shown to the user.

[0003] However, the current scheme for identifying the type of multimedia publisher is not very accurate, and therefore, there is an urgent need for a scheme to solve the above problems. SUMMARY

[0004] To solve or at least partially solve the above problems, the embodiments of the present application provide a method and device for identifying the type of multimedia content publisher.

[0005] In a first aspect, the embodiments of the present application provide a method for identifying the type of multimedia content publisher, the method comprising:

[0006] obtaining information of a to-be-identified multimedia content publisher;

[0007] determining the type of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and a predetermined target association relationship, wherein the target association relationship is used to indicate the association relationship between objects of multiple types, and the objects of multiple types include the multimedia content publisher and a target object associated with the multimedia content publisher.

[0008] Optionally, the target object includes:

[0009] multimedia content, and / or an interactive object.

[0010] Optionally, the determining the type of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and the predetermined target association relationship comprises:

[0011] inputting the information of the to-be-identified multimedia content publisher into a multimedia processing model to obtain the type of the to-be-identified multimedia content publisher, wherein the multimedia processing model is used to determine the type of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and the predetermined target association relationship.

[0012] Optionally, the multimedia processing model includes a first association relationship processing module, and the first association relationship processing module is used to:

[0013] obtaining a feature of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and the predetermined target association relationship;

[0014] obtaining a type of the to-be-identified multimedia content publisher based on the feature of the to-be-identified multimedia content publisher.

[0015] Optionally, the first association relationship processing module is trained by the following manner:

[0016] obtaining a label corresponding to the training multimedia content publisher, the label corresponding to the training multimedia content publisher being used for indicating a type corresponding to the training multimedia content publisher, and the training objects of the plurality of types including the training multimedia content publisher;

[0017] constructing an association relationship between the training objects of the plurality of types based on the training objects of the plurality of types;

[0018] obtaining a fusion feature of the training multimedia content publisher based on the association relationship between the training objects of the plurality of types, and obtaining a predicted type of the training multimedia content based on the fusion feature of the training multimedia content publisher;

[0019] updating parameters of the first association relationship processing module based on the predicted type and the label corresponding to the training multimedia content publisher.

[0020] Optionally, the obtaining of the fusion feature of the training multimedia content publisher based on the association relationship between the training objects of the plurality of types comprises:

[0021] obtaining the fusion feature of the training multimedia content publisher based on an initial feature of the training multimedia content and an initial feature of another training object having an association relationship with the training multimedia content.

[0022] Optionally, the obtaining of the fusion feature of the training multimedia content publisher based on the initial feature of the training multimedia content and the initial feature of another training object having an association relationship with the training multimedia content comprises:

[0023] The fusion feature of the training multimedia content publisher is obtained based on the initial feature of the training multimedia content, the initial features of other training objects having a correlation relationship with the training multimedia content, and attention coefficients of each of the initial features of the other training objects, the other training objects including a first object, and the attention coefficient of the initial feature of the first object being determined based on the initial feature of the first object and the initial feature of the training multimedia content and a correlation relationship type between the training multimedia content and the first object.

[0024] Optionally, the correlation relationship between the plurality of types of training objects includes:

[0025] The correlation relationship between the training multimedia contents includes an initial correlation relationship and an additional correlation relationship obtained by the second correlation relationship processing module based on the features of the training multimedia contents, and the initial correlation relationship is determined based on the training multimedia contents.

[0026] Optionally, the training process of the second correlation relationship processing module includes N rounds of iterations, and the i-th round of iteration is as follows:

[0027] An intermediate correlation relationship of a plurality of training multimedia contents is obtained, and the intermediate correlation relationship includes the initial correlation relationship and additional correlation relationships determined in the previous (i-1) rounds of iterations.

[0028] Features of the plurality of training multimedia contents are obtained based on the intermediate correlation relationship.

[0029] A prediction result of the plurality of training multimedia contents is obtained based on the features of the plurality of training multimedia contents.

[0030] Parameters of the second correlation relationship processing module are updated based on the prediction result of the plurality of training multimedia contents and labels of the plurality of training multimedia contents.

[0031] Optionally, the updating of the parameters of the second correlation relationship processing module based on the prediction result of the plurality of training multimedia contents and the labels of the plurality of training multimedia contents includes:

[0032] The parameters of the second correlation relationship processing module are updated based on the prediction result of the plurality of training multimedia contents, the labels of the plurality of training multimedia contents, and a regularization term of the intermediate correlation relationship.

[0033] Optionally, the obtaining of the plurality of types of training objects includes:

[0034] A plurality of training objects corresponding to each type of the plurality of types are sequentially obtained in the order of the arrangement sequence of the plurality of types.

[0035] Optionally, the method further comprises:

[0036] obtaining multimedia content published by the to-be-identified multimedia content publisher;

[0037] The method further comprises:

[0038] obtaining multimedia content published by the to-be-identified multimedia content publisher;

[0039] Optionally, the multimedia processing model further comprises

[0040] the second association relationship processing module;

[0041] The second association relationship processing module is configured to obtain a target feature of the multimedia content published by the to-be-identified multimedia content publisher based on an association relationship between the multimedia content published by the to-be-identified multimedia content publisher and the training multimedia content;

[0042] The first association relationship processing module is configured to:

[0043] obtain a feature of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher, the target feature, and a predetermined target association relationship;

[0044] obtain a type of the to-be-identified multimedia content publisher based on the feature of the information of the to-be-identified multimedia content publisher.

[0045] Optionally, the method further comprises:

[0046] obtaining information of an interactive object corresponding to the to-be-identified multimedia content publisher;

[0047] The method further comprises:

[0048] obtaining information of an interactive object corresponding to the to-be-identified multimedia content publisher;

[0049] Optionally, the first association relationship processing module is configured to:

[0050] obtain a feature of the information of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher, the information of the interactive object, and the target association relationship;

[0051] obtain the type of the to-be-identified multimedia content publisher based on the feature of the information of the to-be-identified multimedia content publisher.

[0052] In a second aspect, an embodiment of the present application provides a device for identifying a type of a multimedia content publisher, and the device comprises:

[0053] a first obtaining unit, configured to obtain information of a to-be-identified multimedia content publisher;

[0054] a first determining unit, configured to determine the type of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and a target association relationship determined in advance, wherein the target association relationship is used to indicate an association relationship between objects of multiple types, and the objects of the multiple types include a multimedia content publisher and a target object associated with the multimedia content publisher.

[0055] Optionally, the target object includes:

[0056] a multimedia content and / or an interactive object.

[0057] Optionally, the first determining unit is configured to:

[0058] input the information of the to-be-identified multimedia content publisher into a multimedia processing model to obtain the type of the to-be-identified multimedia content publisher, wherein the multimedia processing model is used to determine the type of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and the target association relationship determined in advance.

[0059] Optionally, the multimedia processing model comprises a first association relationship processing module, and the first association relationship processing module is configured to:

[0060] obtain a feature of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and the target association relationship determined in advance;

[0061] obtain the type of the to-be-identified multimedia content publisher based on the feature of the to-be-identified multimedia content publisher.

[0062] Optionally, the first association relationship processing module is obtained by training in the following manner:

[0063] obtaining labels corresponding to the training multimedia content publishers, the labels corresponding to the training multimedia content publishers being used to indicate types corresponding to the training multimedia content publishers, the multiple types of training objects including the training multimedia content publishers;

[0064] constructing association relationships between the multiple types of training objects based on the multiple types of training objects;

[0065] obtaining a fusion feature of the training multimedia content publisher based on the association relationships between the multiple types of training objects, and obtaining a predicted type of the training multimedia content based on the fusion feature of the training multimedia content publisher;

[0066] updating parameters of the first association relationship processing module based on the predicted type and the labels corresponding to the training multimedia content publishers.

[0067] Optionally, the obtaining of the fusion feature of the training multimedia content publisher based on the association relationships between the multiple types of training objects comprises:

[0068] obtaining the fusion feature of the training multimedia content publisher based on an initial feature of the training multimedia content and initial features of other training objects having association relationships with the training multimedia content.

[0069] Optionally, the obtaining of the fusion feature of the training multimedia content publisher based on the initial feature of the training multimedia content and the initial features of the other training objects having the association relationships with the training multimedia content comprises:

[0070] obtaining the fusion feature of the training multimedia content publisher based on the initial feature of the training multimedia content, the initial features of the other training objects having the association relationships with the training multimedia content, and attention coefficients of the initial features of the other training objects, the other training objects including a first object, the attention coefficient of the initial feature of the first object being determined based on the initial feature of the first object, the initial feature of the training multimedia content, and an association relationship type between the training multimedia content and the first object.

[0071] Optionally, the association relationships between the multiple types of training objects comprise:

[0072] association relationships between training multimedia contents, the association relationships between the training multimedia contents including initial association relationships and additional association relationships obtained by a second association relationship processing module based on features of the training multimedia contents, the initial association relationships being determined based on the training multimedia contents.

[0073] Optionally, the training process of the second correlation relationship processing module includes N rounds of iterations, and the i-th round of iteration is as follows:

[0074] obtain intermediate correlation relationships of the plurality of training multimedia contents, the intermediate correlation relationships including the initial correlation relationship and additional correlation relationships determined in previous (i-1) rounds of iterations;

[0075] obtain features of the plurality of training multimedia contents based on the intermediate correlation relationships;

[0076] obtain prediction results of the plurality of training multimedia contents based on the features of the plurality of training multimedia contents;

[0077] update parameters of the second correlation relationship processing module based on the prediction results of the plurality of training multimedia contents and labels of the plurality of training multimedia contents.

[0078] Optionally, the updating of the parameters of the second correlation relationship processing module based on the prediction results of the plurality of training multimedia contents and the labels of the plurality of training multimedia contents includes:

[0079] updating the parameters of the second correlation relationship processing module based on the prediction results of the plurality of training multimedia contents, the labels of the plurality of training multimedia contents, and a regularization term of the intermediate correlation relationships.

[0080] Optionally, the obtaining of the plurality of types of training objects includes:

[0081] obtaining, in sequence, a plurality of training objects corresponding to each type of the plurality of types according to the arrangement order of the plurality of types.

[0082] Optionally, the apparatus further includes:

[0083] a second obtaining unit, configured to obtain multimedia content published by the to-be-identified multimedia content publisher;

[0084] a second determining unit, configured to input information of the to-be-identified multimedia content publisher into a multimedia processing model to obtain a type of the to-be-identified multimedia content publisher, and the inputting includes:

[0085] a third determining unit, configured to input information of the to-be-identified multimedia content publisher and the multimedia content published by the to-be-identified multimedia content publisher into the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher.

[0086] Optionally, the multimedia processing model further includes the second correlation relationship processing module.

[0087] The second association relationship processing module is configured to obtain a target feature of the multimedia content published by the to-be-identified multimedia content publisher based on an association relationship between the multimedia content published by the to-be-identified multimedia content publisher and the training multimedia content.

[0088] The first association relationship processing module is configured to:

[0089] obtain a feature of the information of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher, the target feature, and a target association relationship determined in advance;

[0090] obtain a type of the to-be-identified multimedia content publisher based on the feature of the information of the to-be-identified multimedia content publisher.

[0091] Optionally, the apparatus further includes:

[0092] a third obtaining unit configured to obtain information of an interactive object corresponding to the to-be-identified multimedia content publisher;

[0093] The inputting of the information of the to-be-identified multimedia content publisher into the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher includes:

[0094] inputting the information of the to-be-identified multimedia content publisher and the information of the interactive object into the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher.

[0095] Optionally, the first association relationship processing module is configured to:

[0096] obtain the feature of the information of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher, the information of the interactive object, and the target association relationship;

[0097] obtain the type of the to-be-identified multimedia content publisher based on the feature of the information of the to-be-identified multimedia content publisher.

[0098] In a third aspect, an apparatus is provided, the apparatus including a processor and a memory;

[0099] The processor is configured to execute instructions stored in the memory to cause the apparatus to perform the method of any of the above first aspect.

[0100] In a fourth aspect, a computer-readable storage medium is provided, including instructions instructing an apparatus to perform the method of any of the above first aspect.

[0101] In a fifth aspect, an embodiment of the present application provides a computer program product, which, when executed on a computer, causes the computer to perform the method of any one of the first aspect.

[0102] Compared with the prior art, the embodiments of the present application have the following advantages:

[0103] The embodiments of the present application provide a method for identifying a type of a multimedia content publisher. The method can be executed by a client or a server. In one example, the method comprises: obtaining information of a to-be-identified multimedia content publisher; and determining a type of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and a predetermined target association relationship. The target association relationship is used to indicate an association relationship between objects of multiple types, and the objects of multiple types include a multimedia content publisher and a target object associated with the multimedia content publisher. In the embodiments of the present application, when determining the type of the to-be-identified multimedia content publisher, the association relationship between objects of multiple types is considered in addition to the information of the to-be-identified multimedia content publisher, for example, the association relationship between multimedia content publishers and the association relationship between a multimedia content publisher and a target object associated with the multimedia content publisher. Therefore, the type of the to-be-identified multimedia content publisher can be accurately determined by using the solution of the embodiments of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0104] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0105] Figure 1 A flowchart of a method for identifying a type of a multimedia content publisher provided by an embodiment of the present application is shown in FIG. 1.

[0106] Figure 2 A flowchart of a training method of a first association relationship processing module provided by an embodiment of the present application is shown in FIG. 2.

[0107] Figure 3 A flowchart of an i-th round of iteration in a process of training a second association relationship processing module provided by an embodiment of the present application is shown in FIG. 3.

[0108] Figure 4 A structural flowchart of a device for identifying a type of a multimedia content publisher provided by an embodiment of the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0109] In order to better understand the technical scheme of the present application, the technical scheme of the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0110] The inventors of the present application have found that at present, a decision tree model can be trained to identify the type of a multimedia content publisher. For example, a training multimedia content publisher and a label of the training multimedia content publisher can be obtained to train the aforementioned decision tree model, wherein the label of the training multimedia content publisher is used to indicate the type of the training multimedia content publisher. However, when the decision tree model is used to determine the type of a to-be-identified multimedia content publisher, the information referred to only includes the information of the to-be-identified multimedia content publisher, and the information referred to is limited. Accordingly, the accuracy of the identification result is not very high.

[0111] In order to solve the above problems, the embodiments of the present application provide a method and device for identifying the type of a multimedia content publisher.

[0112] The various non-limiting embodiments of the present application will be described in detail below in conjunction with the drawings.

[0113] Exemplary method

[0114] Referring to Figure 1 The figure is a flowchart of a method for identifying the type of a multimedia content publisher provided by the embodiments of the present application. In the present embodiment, the method may, for example, include the following steps: S101-S102.

[0115] Before introducing the method of the embodiments of the present application, the terms involved in the embodiments of the present application will be described first.

[0116] Multimedia content: including but not limited to one or more of audio, video, image, character (such as text). In one example, the multimedia content can be an advertisement.

[0117] Interactive object: can be an object interacting with multimedia content. The object interacting with multimedia content can be a user interacting with multimedia content. Interacting with multimedia content includes but is not limited to browsing, clicking, commenting on the multimedia content.

[0118] Multimedia content publisher: a publisher of multimedia content. In one example, when the multimedia content is an advertisement, the multimedia content publisher can be an advertiser.

[0119] S101: Obtain information of a multimedia content publisher to be identified.

[0120] In one example, the multimedia content publisher to be identified can be a multimedia content publisher that is about to publish multimedia content. In one example, before the multimedia content publisher to be identified publishes multimedia content, information of the multimedia content publisher to be identified can be obtained, the multimedia content publisher to be identified can be identified by type, and it can be determined whether the multimedia content publisher to be identified is allowed to publish multimedia content. In another example, the type of the multimedia content publisher to be identified can be displayed when the multimedia content published by the multimedia content publisher to be identified is played.

[0121] In one example, the type of the multimedia content publisher to be identified can include normal or abnormal.

[0122] In the embodiments of the present application, the information of the multimedia content publisher to be identified can be information related to the multimedia content publisher to be identified. The information of the multimedia content publisher to be identified can include attribute information of the multimedia content publisher to be identified, which includes but is not limited to the field to which it belongs, name, and the like.

[0123] S102: Determine the type of the multimedia content publisher to be identified based on the information of the multimedia content publisher to be identified and a predetermined target association relationship. The target association relationship is used to indicate an association relationship between objects of multiple types, and the objects of multiple types include a multimedia content publisher and a target object associated with the multimedia content publisher.

[0124] After the multimedia content to be identified is obtained, the type of the multimedia content publisher to be identified can be determined based on the information of the multimedia content publisher to be identified and a predetermined target association relationship.

[0125] Regarding the target association relationship, it should be noted that the target association relationship can be used to indicate an association relationship between objects of multiple types. In addition to including a multimedia content publisher, the objects of multiple types can also include a target object associated with the multimedia content publisher. In other words, the target association relationship can be used to indicate an association relationship between multimedia content publishers, an association relationship between a multimedia content publisher and a target object, and an association relationship between target objects.

[0126] The target object is not specifically limited in the embodiments of the present application, and can be any object having an association relationship with the multimedia content publisher. In one example, considering that the multimedia content publisher can be used to publish multimedia content, the association relationship between the multimedia content and the multimedia content publisher is relatively close, and therefore the target object can include the multimedia content. In another example, considering that the multimedia content publisher and the interactive object can have an association relationship through the multimedia content published by the multimedia content publisher, the target object can include the interactive object.

[0127] In one example, the target association relationship can be embodied as a target association relationship graph, which includes multiple types of nodes, one node corresponding to one object. For example, one multimedia content publisher corresponds to one node, one multimedia content corresponds to one node, and one interactive object corresponds to one node. The association relationship between objects can be embodied as an edge between nodes, for example, nodes A and B have an association relationship, and therefore there is an edge between the nodes A and B.

[0128] Based on the information of the to-be-identified multimedia content publisher and the target association relationship, the type of the to-be-identified multimedia content publisher is determined, so that in determining the type of the to-be-identified multimedia content publisher, in addition to considering the information of the to-be-identified multimedia content publisher itself, the association relationship between multiple types of objects is also considered, for example, the association relationship between multimedia content publishers, the association relationship between the multimedia content publisher and the target object associated with the multimedia content publisher, and the association relationship between the target objects. Therefore, the type of the to-be-identified multimedia content publisher can be accurately determined.

[0129] In one example, S102, when specifically implemented, can use a specific algorithm to establish a connection between the to-be-identified multimedia content publisher and the target association relationship, thereby further determining the type of the to-be-identified multimedia content.

[0130] In another example, S102, when specifically implemented, can be implemented by means of a multimedia processing model. The multimedia processing model is used to determine the type of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and the pre-determined target association relationship. For this case, S102, when specifically implemented, can input the information of the to-be-identified multimedia content publisher into the multimedia processing model, and accordingly, the multimedia processing model can output the type of the to-be-identified multimedia content publisher.

[0131] In one example, the aforementioned "determining the type of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and the pre-determined target association relationship" can be implemented by a first association relationship processing module in the multimedia processing model. Specifically, after inputting the information of the to-be-identified multimedia content publisher into the multimedia processing model, the first association relationship processing module can determine the type of the to-be-identified multimedia content publisher by the following steps A1-A2.

[0132] Step A1: obtaining the feature of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher and the pre-determined target association relationship.

[0133] In one example, the first association relationship processing module can establish an association relationship between the to-be-identified multimedia content publisher and the target association relationship based on the information of the to-be-identified multimedia content publisher, for example, obtaining a first association relationship. The first association relationship can be used to indicate the association relationship between the to-be-identified multimedia content publisher and other objects in the target association relationship. Then, the feature of the to-be-identified multimedia content publisher can be obtained based on the first association relationship. For example, the feature of the to-be-identified multimedia content publisher can be obtained based on the initial feature of the to-be-identified multimedia content publisher and the features of other objects having an association relationship with the to-be-identified multimedia content publisher in the first association relationship. In one example, the target association relationship can be an association relationship pre-stored by the first association relationship processing module.

[0134] Step A2: obtaining the type of the to-be-identified multimedia content publisher based on the feature of the to-be-identified multimedia content publisher.

[0135] After obtaining the feature of the to-be-identified multimedia content, the type of the to-be-identified multimedia content publisher can be obtained based on the feature of the to-be-identified multimedia content publisher. For example, the first association relationship processing module can include at least one fully connected layer, and the at least one fully connected layer can be used to analyze the feature of the to-be-identified multimedia content publisher to obtain the type of the to-be-identified multimedia content publisher.

[0136] In one example, the first association relationship processing module can be pre-trained, and for this case, the target association relationship can be obtained after the training of the target association relationship is completed.

[0137] Next, the method for training the first association relationship processing module will be introduced. Figure 2 Figure 2 ​A flowchart of a training method of a first association relationship processing module is provided in the embodiments of the present application. Figure 2 The method shown may, for example, include steps S201-S204.

[0138] S201: Obtain labels corresponding to the training multimedia content publishers of the multiple types of training objects, the labels corresponding to the training multimedia content publishers being used to indicate the types corresponding to the training multimedia content publishers, and the multiple types of training objects including the training multimedia content publishers.

[0139] In one example, the training objects include training multimedia content publishers, training multimedia content, and training interactive objects. The training multimedia content publishers may be multimedia content publishers that have published multimedia content, the training multimedia content may be multimedia content published by the training multimedia content publishers, and the training interactive objects may be objects that interact with the training multimedia content.

[0140] After the multiple types of training objects are obtained, the first association relationship processing module can be trained using the multiple types of training objects.

[0141] In one example, considering the process of training the first association relationship processing module, the process may include the process of learning the association relationships between the objects by the first association relationship processing module, or in other words, the process of updating the association relationship graph between the objects. In one example, in order to make the association relationships between the objects of the same type have better connectivity, S201 may, in specific implementation, obtain multiple training objects corresponding to each type of the multiple types in sequence according to the arrangement order of the multiple types. For example, multiple training multimedia content publishers are first obtained to train the first association relationship processing module, then multiple training multimedia content are obtained to train the first association relationship processing module, and finally multiple training interactive objects are obtained to train the first association relationship processing module, in the order of multimedia content publishers, multimedia content, and interactive objects.

[0142] In the embodiments of the present application, obtaining the multiple types of training objects refers to obtaining information of the multiple types of training objects.

[0143] S202: Construct the association relationships between the multiple types of training objects based on the multiple types of training objects.

[0144] In one example, constructing the association relationships between the multiple types of training objects based on the multiple types of training objects may be constructing the association relationships between the multiple types of training objects based on information of the multiple types of training objects.

[0145] In one example, the association relationship between the multiple types of training objects can be constructed based on preset rules. For example, training multimedia content publishers belonging to the same organization (e.g., an enterprise or a group) have an association relationship, a training multimedia content publisher and the training multimedia content published by it have an association relationship, an association relationship between training multimedia contents is determined based on the content features of the to-be-identified multimedia content, and an interactive object interacting with the training multimedia content has an association relationship with the training multimedia content.

[0146] Regarding the association relationship between the training multimedia contents, it should be noted that:

[0147] In one example, for each training multimedia content, the text features and video features of the training multimedia content can be extracted respectively, and the text features and video features can be fused to obtain the fusion features of the training multimedia content. Further, based on the fusion features of each training multimedia content, the association relationship between the training multimedia contents (hereinafter referred to as an initial association relationship) is obtained. For example, the initial association relationship can be obtained by using a K-Nearest Neighbor (KNN) algorithm. As an example, the text features of the training multimedia content can be extracted by using bert, and the video features of the training multimedia content can be extracted by using an Unsupervised teacher-student (UTS).

[0148] In yet another example, considering that the aforementioned initial association relationship determined based on the training multimedia content is relatively simple, therefore, as an example, the second association relationship processing module can also be used to determine additional association relationships between the training multimedia contents based on the features of the training multimedia contents. Accordingly, for this case, the association relationship between the aforementioned training multimedia contents can include the aforementioned initial association relationship and the additional association relationship.

[0149] Regarding the second association relationship processing module, it can be obtained by N rounds of iterative training in advance, and the process of the i-th round of iteration can refer to the description of the second association relationship processing module in the following part, which will not be described in detail here. Figure 3

[0150] S203: Based on the association relationship between the multiple types of training objects, the fusion features of the training multimedia content publisher are obtained, and based on the fusion features of the training multimedia content publisher, the predicted type of the training multimedia content is obtained.

[0151] ​In one example, for any training multimedia content publisher, its fusion feature can be obtained according to the initial features of other training objects having a correlation relationship with it. For example, the initial features of other objects having a correlation relationship with the training multimedia content publisher can be fused according to a preset fusion manner to obtain the fusion feature of the training multimedia content publisher.

[0152] Regarding the initial feature of a training object, in one example, the initial feature of the training object can include a content feature of the training object and an object type feature of the training object. The content feature of the training object and the object type feature of the training object can be spliced in dimension to obtain the initial feature of the training object. Regarding the object type feature, it should be noted that, in one example, when the type of a training object for training the first correlation relationship processing module is N, the training object feature can be an N-dimensional feature. For example, the type of a training object includes a training multimedia content publisher, a training multimedia content, and a training interactive object, and the training object feature can be a 3-dimensional feature. For example, the training object feature of a training multimedia content publisher is 100, the training object feature of a training multimedia content is 010, and the training object feature of a training interactive object is 001.

[0153] In one example, considering that the correlation degrees between each training object and the training multimedia content publisher are different among multiple training objects having a correlation relationship with the training multimedia content publisher. Therefore, in one example, based on the initial feature of the training multimedia content and the initial features of other training objects having a correlation relationship with the training multimedia content, the fusion feature of the training multimedia content publisher can also be combined with attention coefficients of multiple training objects having a correlation relationship with the training multimedia content in specific implementation. Specifically:

[0154] The fusion feature of the training multimedia content publisher can be obtained based on the initial feature of the training multimedia content, the initial features of other training objects having a correlation relationship with the training multimedia content, and the attention coefficients of each initial feature in the initial features of the other training objects.

[0155] The attention coefficient of the other training object is described by taking any one of the other training objects as an example. For convenience of description, any one of the other training objects is referred to as a first object. In an example, the attention coefficient of the initial feature of the first object is determined based on the initial feature of the first object, the initial feature of the training multimedia content, and the association type between the training multimedia content and the first object. For example, the association type between the training multimedia content and the first object can be a homogeneous association type, or can be a non-homogeneous association type, where:

[0156] The homogeneous association type can be understood as a type corresponding to an association between objects of the same type. For example, the association type between training multimedia contents is a homogeneous association type.

[0157] The non-homogeneous association type can be understood as a type corresponding to an association between objects of different types. For example, the association type between the training multimedia content and the training multimedia content publisher is a non-homogeneous association type. For another example, the association type between the training multimedia content and the training interactive object is a non-homogeneous association type.

[0158] After determining the fusion feature, the predicted type of the training multimedia content can be obtained based on the fusion feature of the training multimedia content. For example, the first association relationship processing module includes a plurality of fully connected layers, and the fusion feature of the training multimedia content can be processed by using the plurality of fully connected layers to obtain the predicted type of the training multimedia content.

[0159] S204: Update the parameters of the first association relationship processing module based on the predicted type and the label corresponding to the training multimedia content publisher.

[0160] In an example, after obtaining the predicted type, a first loss function can be calculated based on the predicted type and the label of the training multimedia content publisher, and the parameters of the first association relationship processing module can be updated based on the first loss function.

[0161] Next, the training of the second association relationship processing module is described in combination with Figure 3 The method for training the second association relationship processing module is described. Figure 3 A flowchart of a process for training the second association relationship processing module in the i th iteration is provided.

[0162] In an example, the second association relationship processing module is trained, and then the association relationship between the training multimedia contents is obtained. In an example, the association relationship between the training multimedia contents can be embodied as an association relationship graph between the training multimedia contents, in which one training multimedia content corresponds to one node, and an edge between nodes is used to indicate the association relationship between the nodes.

[0163] Figure 3 The method shown can include the following steps S301-S304, for example.

[0164] S301: Obtain the intermediate association relationship of the plurality of training multimedia contents, the intermediate association relationship including the initial association relationship and the additional association relationship determined in the previous (i-1) rounds of iterations.

[0165] Here, i is an integer greater than or equal to 1, when the i is equal to 1, the intermediate association relationship is the initial association relationship, and the initial association relationship can be referred to the relevant description part above, which will not be described in detail this time.

[0166] S302: Obtain the features of the plurality of training multimedia contents based on the intermediate association relationship.

[0167] For each training multimedia content, its features can be obtained based on its own content features and the content features of other training multimedia objects having an association relationship with it. For example, the content features of the training multimedia content itself and the content features of other training multimedia objects having an association relationship with it can be fused to obtain the features of the training multimedia content.

[0168] In an example, the intermediate association relationship can be an intermediate association relationship graph.

[0169] S303: Obtain the prediction results of the plurality of training multimedia contents based on the features of the plurality of training multimedia contents.

[0170] For each training multimedia content, after obtaining the features of the training multimedia content, the prediction result of the training multimedia content can be obtained based on the features of the training multimedia content. For example, the second association relationship processing module can include at least one fully connected layer, and the features of the training multimedia content can be processed by using the at least one fully connected layer to obtain the prediction result of the training multimedia content.

[0171] S304: Update the parameters of the second association relationship processing module based on the prediction results of the plurality of training multimedia contents and the labels of the plurality of training multimedia contents.

[0172] In one example, after obtaining the prediction result, a second loss function can be calculated based on the prediction result and the label of the training multimedia content, so as to update the parameters of the second association processing module based on the second loss function. As an example, the second loss function can be a cross-entropy function.

[0173] In yet another example, in order to make the association graph between the training multimedia content obtained by the trained second association processing module more smooth, the parameters of the second association processing module can also be updated in combination with the regular term of the intermediate association. For example, the parameters of the second association processing module can be updated based on the aforementioned second loss function and the regular term.

[0174] In the embodiments of the present application, when training the second association processing module, the label of the to-be-recognized multimedia content is used to optimize the parameters of the second association processing module, so that the association between the training multimedia content learned by the second association processing module includes the association between the labels of the training multimedia content. Therefore, the additional association between the training multimedia content obtained by using the second association processing module at least includes the association in the label dimension of the training multimedia content.

[0175] In one example, in order to improve the accuracy of determining the type of the to-be-recognized multimedia content publisher, in addition to obtaining the information of the to-be-recognized multimedia content publisher, the multimedia content published by the to-be-recognized multimedia content publisher can also be obtained, and further, the type of the to-be-recognized multimedia content publisher is determined based on the information of the to-be-recognized multimedia content publisher and the multimedia content published by the to-be-recognized multimedia content publisher. As an example, the information of the to-be-recognized multimedia content publisher and the multimedia content published by the to-be-recognized multimedia content publisher can be input into the multimedia processing model to obtain the type of the to-be-recognized multimedia content publisher.

[0176] For this case, after inputting the information of the to-be-recognized multimedia content publisher and the multimedia content published by the to-be-recognized multimedia content publisher into the multimedia processing model, the multimedia processing model can analyze the information of the to-be-recognized multimedia content publisher and the multimedia content published by the to-be-recognized multimedia content publisher to obtain the type of the to-be-recognized multimedia content publisher.

[0177] In one example, in addition to the aforementioned first association processing module, the multimedia processing model can also include the second association processing module. For this case:

[0178] The second association relationship processing module can obtain a target feature of the multimedia content published by the to-be-identified multimedia content publisher based on an association relationship between the multimedia content published by the to-be-identified multimedia content publisher and the training multimedia content. For example, the second association relationship processing module can add the multimedia content published by the to-be-identified multimedia content publisher into the association relationship between the training multimedia content based on a content feature of the multimedia content published by the to-be-identified multimedia content publisher, and obtain a second association relationship. Further, the target feature of the multimedia content published by the to-be-identified multimedia content publisher is determined based on the second association relationship. For example, the content feature of the multimedia content published by the to-be-identified multimedia content publisher and other associated features can be fused to obtain the target feature. The other associated features are features of the training multimedia content indicated by the second association relationship and having an association relationship with the multimedia content published by the to-be-identified multimedia content publisher.

[0179] The first association relationship processing module can be configured to perform the following steps B1-B2 to obtain the type of the to-be-identified multimedia content publisher.

[0180] Step B1: obtaining a feature of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher, the target feature, and a predetermined target association relationship.

[0181] In one example, the first association relationship processing module can establish an association relationship between a third association relationship and a target association relationship based on the information of the to-be-identified multimedia content publisher and the target feature, and obtain a fourth association relationship. The third association relationship is an association relationship between the to-be-identified multimedia content publisher and the multimedia content published by the to-be-identified multimedia content publisher. Then, the feature of the to-be-identified multimedia content publisher can be obtained based on the fourth association relationship. For example, the feature of the to-be-identified multimedia content publisher can be obtained based on an initial feature of the to-be-identified multimedia content publisher and features of other objects having an association relationship with the to-be-identified multimedia content publisher in the fourth association relationship.

[0182] Step B2: obtaining the type of the to-be-identified multimedia content publisher based on the feature of the to-be-identified multimedia content publisher.

[0183] For the step B2, refer to the description of the step A2 above, which will not be described in detail here.

[0184] In one example, in order to improve the accuracy of determining the type of the to-be-identified multimedia content publisher, in addition to obtaining the information of the to-be-identified multimedia content publisher, the information of the interaction object corresponding to the to-be-identified multimedia content publisher can also be obtained. The interaction object corresponding to the to-be-identified multimedia content publisher mentioned here can be, for example, an interaction object that interacts with the multimedia content that has been historically published by the to-be-identified multimedia content publisher.

[0185] Further, based on the information of the to-be-identified multimedia content publisher and the information of the interaction object corresponding to the to-be-identified multimedia content publisher, the type of the to-be-identified multimedia content publisher is determined. As an example, the information of the to-be-identified multimedia content publisher and the information of the interaction object corresponding to the to-be-identified multimedia content can be input into the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher.

[0186] For this case, after inputting the information of the to-be-identified multimedia content publisher and the information of the interaction object corresponding to the to-be-identified multimedia content publisher into the multimedia processing model, the multimedia processing model can analyze the information of the to-be-identified multimedia content publisher and the information of the interaction object corresponding to the to-be-identified multimedia content publisher to obtain the type of the to-be-identified multimedia content publisher.

[0187] In one example, the information of the to-be-identified multimedia content publisher and the information of the interaction object corresponding to the to-be-identified multimedia content publisher can be analyzed by a first association relationship processing module included in the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher. As an example, the first association relationship processing module can perform the following steps C1-C2 to determine the type of the to-be-identified multimedia content publisher.

[0188] Step C1: based on the information of the to-be-identified multimedia content publisher, the information of the interaction object, and the target association relationship, a feature of the information of the to-be-identified multimedia content publisher is obtained.

[0189] In one example, the first association relationship processing module can establish an association relationship between the fifth association relationship and the target association relationship based on the information of the to-be-identified multimedia content publisher and the information of the interactive object. For example, a sixth association relationship is obtained. The fifth association relationship is an association relationship between the to-be-identified multimedia content publisher and the interactive object corresponding to the to-be-identified multimedia content publisher. Then, the feature of the to-be-identified multimedia content publisher can be obtained based on the sixth association relationship. For example, the feature of the to-be-identified multimedia content publisher can be obtained based on the initial feature of the to-be-identified multimedia content publisher and the features of other objects having an association relationship with the to-be-identified multimedia content publisher in the sixth association relationship.

[0190] Step C2: obtaining the type of the to-be-identified multimedia content publisher based on the feature of the information of the to-be-identified multimedia content publisher.

[0191] For the step C2, reference can be made to the description of the step A2 above, which will not be described in detail herein. Of course, in another example, the information of the to-be-identified multimedia content publisher, the multimedia content published by the to-be-identified multimedia content publisher, and the information of the interactive object corresponding to the to-be-identified multimedia content publisher can all be input into the multimedia processing model, so as to obtain the type of the to-be-identified multimedia content. For this case, the second association relationship learning module is configured to obtain a target feature of the multimedia content published by the to-be-identified multimedia content publisher based on the association relationship between the multimedia content published by the to-be-identified multimedia content publisher and the training multimedia content. The first association relationship processing module is configured to obtain the feature of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher, the target feature, the information of the interactive object corresponding to the to-be-identified multimedia content publisher, and the target association relationship determined in advance, and further obtain the type of the to-be-identified multimedia content publisher based on the feature of the information of the to-be-identified multimedia content publisher. The operations performed by the first association relationship processing module and the operations performed by the second association relationship processing module can refer to the related description above, which will not be described herein.

[0192] Exemplary device

[0193] Based on the method provided in the above examples, the embodiments of the present application further provide an apparatus, which will be described below with reference to the accompanying drawings.

[0194] Reference is made to Figure 4The figure is a structural flow diagram of an apparatus for identifying a type of a multimedia content publisher according to an embodiment of the present application. The apparatus 400 may, for example, specifically include a first obtaining unit 401 and a first determining unit 402.

[0195] The first obtaining unit 401 is configured to obtain information of a multimedia content publisher to be identified.

[0196] The first determining unit 402 is configured to determine a type of the multimedia content publisher to be identified based on the information of the multimedia content publisher to be identified and a predetermined target association relationship. The target association relationship is used to indicate an association relationship between a plurality of types of objects, and the plurality of types of objects include a multimedia content publisher and a target object associated with the multimedia content publisher.

[0197] Optionally, the target object includes:

[0198] A multimedia content and / or an interactive object.

[0199] Optionally, the first determining unit 402 is configured to:

[0200] input the information of the multimedia content publisher to be identified into a multimedia processing model to obtain the type of the multimedia content publisher to be identified. The multimedia processing model is used to determine the type of the multimedia content publisher to be identified based on the information of the multimedia content publisher to be identified and the predetermined target association relationship.

[0201] Optionally, the multimedia processing model includes a first association relationship processing module. The first association relationship processing module is configured to:

[0202] obtain a feature of the multimedia content publisher to be identified based on the information of the multimedia content publisher to be identified and the predetermined target association relationship;

[0203] obtain the type of the multimedia content publisher to be identified based on the feature of the multimedia content publisher to be identified.

[0204] Optionally, the first association relationship processing module is obtained by training in the following manner:

[0205] obtain labels corresponding to a plurality of types of training objects and training multimedia content publishers. The labels corresponding to the training multimedia content publishers are used to indicate types corresponding to the training multimedia content publishers. The plurality of types of training objects include the training multimedia content publishers.

[0206] construct an association relationship between the plurality of types of training objects based on the plurality of types of training objects.

[0207] obtain a fusion feature of the training multimedia content publisher based on the association relationship between the plurality of types of training objects, and obtain a predicted type of the training multimedia content based on the fusion feature of the training multimedia content publisher;

[0208] update parameters of the first association relationship processing module based on the predicted type and a label corresponding to the training multimedia content publisher.

[0209] Optionally, the obtaining of the fusion feature of the training multimedia content publisher based on the association relationship between the plurality of types of training objects comprises:

[0210] obtaining the fusion feature of the training multimedia content publisher based on the initial feature of the training multimedia content and initial features of other training objects having an association relationship with the training multimedia content.

[0211] Optionally, the obtaining of the fusion feature of the training multimedia content publisher based on the initial feature of the training multimedia content and initial features of other training objects having an association relationship with the training multimedia content comprises:

[0212] obtaining the fusion feature of the training multimedia content publisher based on the initial feature of the training multimedia content, initial features of other training objects having an association relationship with the training multimedia content, and attention coefficients of each initial feature, the other training objects including a first object, the attention coefficient of the initial feature of the first object being determined based on the initial feature of the first object, and the initial feature of the training multimedia content and an association relationship type between the training multimedia content and the first object.

[0213] Optionally, the association relationship between the plurality of types of training objects comprises:

[0214] an association relationship between training multimedia contents, the association relationship between the training multimedia contents comprising an initial association relationship and an additional association relationship obtained by the second association relationship processing module based on features of the training multimedia contents, the initial association relationship being determined based on the training multimedia contents.

[0215] Optionally, the training process of the second association relationship processing module comprises N rounds of iterations, and the i-th round of iteration is as follows:

[0216] obtaining an intermediate association relationship of the plurality of training multimedia contents, the intermediate association relationship comprising the initial association relationship and additional association relationships determined in the previous (i-1) rounds of iterations;

[0217] obtaining features of the plurality of training multimedia contents based on the intermediate association relationship;

[0218] obtaining prediction results of the plurality of training multimedia contents based on the features of the plurality of training multimedia contents;

[0219] updating parameters of the second association relationship processing module based on the prediction results of the plurality of training multimedia contents and labels of the plurality of training multimedia contents.

[0220] Optionally, the updating the parameters of the second association relationship processing module based on the prediction results of the plurality of training multimedia contents and the labels of the plurality of training multimedia contents comprises:

[0221] updating the parameters of the second association relationship processing module based on the prediction results of the plurality of training multimedia contents, the labels of the plurality of training multimedia contents, and a regularization term of the intermediate association relationship.

[0222] Optionally, the obtaining the plurality of types of training objects comprises:

[0223] obtaining, in sequence, the plurality of training objects corresponding to each type of the plurality of types according to the arrangement order of the plurality of types.

[0224] Optionally, the apparatus further comprises:

[0225] a second obtaining unit, configured to obtain multimedia content published by the to-be-identified multimedia content publisher;

[0226] a second determining unit, configured to input information of the to-be-identified multimedia content publisher into a multimedia processing model to obtain a type of the to-be-identified multimedia content publisher, and the inputting the information of the to-be-identified multimedia content publisher into the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher comprises:

[0227] a third determining unit, configured to input the information of the to-be-identified multimedia content publisher and the multimedia content published by the to-be-identified multimedia content publisher into the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher.

[0228] Optionally, the multimedia processing model further comprises the second association relationship processing module;

[0229] the second association relationship processing module is configured to obtain target features of the multimedia content published by the to-be-identified multimedia content publisher based on an association relationship between the multimedia content published by the to-be-identified multimedia content publisher and the training multimedia content;

[0230] the first association relationship processing module is configured to:

[0231] obtaining a feature of the information of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher, the target feature, and the predetermined target association relationship;

[0232] obtaining a type of the to-be-identified multimedia content publisher based on the feature of the information of the to-be-identified multimedia content publisher.

[0233] Optionally, the apparatus further includes:

[0234] a third obtaining unit, configured to obtain information of an interactive object corresponding to the to-be-identified multimedia content publisher;

[0235] the inputting of the information of the to-be-identified multimedia content publisher into the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher includes:

[0236] the inputting of the information of the to-be-identified multimedia content publisher and the information of the interactive object into the multimedia processing model to obtain the type of the to-be-identified multimedia content publisher.

[0237] Optionally, the first association relationship processing module is configured to:

[0238] obtain a feature of the information of the to-be-identified multimedia content publisher based on the information of the to-be-identified multimedia content publisher, the information of the interactive object, and the target association relationship;

[0239] obtain a type of the to-be-identified multimedia content publisher based on the feature of the information of the to-be-identified multimedia content publisher.

[0240] Since the apparatus 400 is a device corresponding to the method provided in the above method embodiments, the specific implementation of each unit of the apparatus 400 is of the same concept as the above method embodiments, and therefore, the specific implementation of each unit of the apparatus 400 can be referred to the description of the above method embodiments, which will not be repeated here.

[0241] The present embodiment also provides a device including a processor and a memory;

[0242] The processor is configured to execute instructions stored in the memory, so that the device performs the method for identifying a type of a multimedia content publisher provided in the above method embodiments.

[0243] The present embodiment provides a computer readable storage medium including instructions, which instruct a device to perform the method for identifying a type of a multimedia content publisher provided in the above method embodiments.

[0244] The embodiment of the present application further provides a computer program product, which, when running on a computer, causes the computer to execute the method for identifying the type of a multimedia content publisher provided in the above method embodiment.

[0245] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with the true scope and spirit of the application being indicated by the following claims.

[0246] It will be understood that the application is not limited to the precise structures hereinbefore described and illustrated in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the application is indicated by the appended claims, and all changes that come within the meaning and range of equivalents are intended to be embraced therein.

[0247] The above description is merely illustrative of the application, and not restrictive. Since the application can be modified in various ways and replaced with equivalent arrangements without departing from the application's spirit and principles, it should not be limited to the scope of the application described above, and all modifications, equivalent replacements, improvements, etc. made within the scope of the application should be included in the scope of the application.

Claims

1. A method for identifying the type of multimedia content publisher, characterized in that, The method includes: Obtain information about the publisher of the multimedia content to be identified; The information of the multimedia content publisher to be identified is input into a multimedia processing model to obtain the type of the multimedia content publisher to be identified. The multimedia processing model includes a first association processing module, which is used to determine the type of the multimedia content publisher to be identified based on the information of the multimedia content publisher to be identified and a pre-determined target association relationship. The target association relationship indicates the association relationship between multiple types of objects, including: multimedia content publishers and target objects associated with the multimedia content publishers. The parameters of the first association processing module are updated based on the association relationships between various types of training objects, including: The association between training multimedia content includes an initial association and an additional association obtained by a second association processing module based on the features of the training multimedia content. The initial association is determined based on the training multimedia content.

2. The method according to claim 1, characterized in that, The target object includes: Multimedia content, and / or interactive objects.

3. The method according to claim 1, characterized in that, The first association processing module is specifically used for: Based on the information of the multimedia content publisher to be identified and the pre-determined target association, the characteristics of the multimedia content publisher to be identified are obtained; Based on the characteristics of the multimedia content publisher to be identified, the type of the multimedia content publisher to be identified is obtained.

4. The method according to claim 3, characterized in that, The first relationship processing module is trained in the following way: Obtain the tags corresponding to the various types of training objects and training multimedia content publishers. The tags corresponding to the training multimedia content publishers are used to indicate the type of the training multimedia content publishers. The various types of training objects include the training multimedia content publishers. Based on the various types of training objects, construct the association relationships between the various types of training objects; Based on the correlation between the various types of training objects, the fusion features of the training multimedia content publisher are obtained, and based on the fusion features of the training multimedia content publisher, the predicted type of the training multimedia content is obtained. Based on the prediction type and the tags corresponding to the trained multimedia content publishers, the parameters of the first association processing module are updated.

5. The method according to claim 4, characterized in that, The fusion features of the trained multimedia content publisher are obtained based on the association relationships between the various types of training objects, including: Based on the initial features of the training multimedia content and the initial features of other training objects that are related to the training multimedia content, the fusion features of the training multimedia content publisher are obtained.

6. The method according to claim 5, characterized in that, The process of obtaining the fused features of the training multimedia content publisher based on the initial features of the training multimedia content and the initial features of other training objects that are related to the training multimedia content includes: Based on the initial features of the training multimedia content, the initial features of other training objects that are related to the training multimedia content, and the attention coefficients of each initial feature in the initial features of the other training objects, the fusion features of the training multimedia content publisher are obtained. The other training objects include a first object. The attention coefficients of the initial features of the first object are determined based on the initial features of the first object, the initial features of the training multimedia content, and the type of association between the training multimedia content and the first object.

7. The method according to claim 1, characterized in that, The training process of the second association processing module includes N rounds of iteration, and the i-th round of iteration is as follows: Obtain intermediate associations among multiple training multimedia contents, wherein the intermediate associations include the initial associations and the additional associations determined in the first (i-1) iterations; Based on the aforementioned intermediate association, the features of the multiple training multimedia contents are obtained; Based on the features of the multiple training multimedia contents, the prediction results of the multiple training multimedia contents are obtained. Based on the prediction results and labels of the multiple training multimedia contents, the parameters of the second association processing module are updated.

8. The method according to claim 7, characterized in that, The step of updating the parameters of the second association processing module based on the prediction results and labels of the multiple training multimedia contents includes: Based on the prediction results of the multiple training multimedia contents, the labels of the multiple training multimedia contents, and the regularization terms of the intermediate association relationships, the parameters of the second association relationship processing module are updated.

9. The method according to claim 4, characterized in that, The acquisition of the various types of training objects includes: According to the arrangement order of the various types, obtain multiple training objects corresponding to each type in sequence.

10. The method according to claim 1, characterized in that, The method further includes: Obtain the multimedia content published by the publisher of the multimedia content to be identified; The step of inputting the information of the multimedia content publisher to be identified into the multimedia processing model to obtain the type of the multimedia content publisher to be identified includes: The information of the multimedia content publisher to be identified and the multimedia content published by the multimedia content publisher to be identified are input into the multimedia processing model to obtain the type of the multimedia content publisher to be identified.

11. The method according to claim 10, characterized in that, The multimedia processing model also includes The second association processing module; The second association processing module is used to obtain the target features of the multimedia content published by the multimedia content publisher to be identified based on the association between the multimedia content published by the multimedia content publisher to be identified and the training multimedia content. The first association processing module is used for: Based on the information of the multimedia content publisher to be identified, the target features, and the pre-determined target associations, the features of the multimedia content publisher to be identified are obtained; Based on the characteristics of the information of the multimedia content publisher to be identified, the type of the multimedia content publisher to be identified is obtained.

12. The method according to claim 1, characterized in that, The method further includes: Obtain information about the interactive object corresponding to the publisher of the multimedia content to be identified; The step of inputting the information of the multimedia content publisher to be identified into the multimedia processing model to obtain the type of the multimedia content publisher to be identified includes: The information of the multimedia content publisher to be identified and the information of the interactive object are input into the multimedia processing model to obtain the type of the multimedia content publisher to be identified.

13. The method according to claim 12, characterized in that, The first association processing module is used for: Based on the information of the multimedia content publisher to be identified, the information of the interactive object, and the target association, the characteristics of the information of the multimedia content publisher to be identified are obtained; Based on the characteristics of the information of the multimedia content publisher to be identified, the type of the multimedia content publisher to be identified is obtained.

14. A device for identifying the type of multimedia content publisher, characterized in that, The device includes: The first acquisition unit is used to acquire information about the publisher of the multimedia content to be identified. A first determining unit is configured to input the information of the multimedia content publisher to be identified into a multimedia processing model to obtain the type of the multimedia content publisher to be identified. The multimedia processing model includes a first association processing module, which is configured to determine the type of the multimedia content publisher to be identified based on the information of the multimedia content publisher to be identified and a pre-determined target association relationship. The target association relationship indicates the association relationship between multiple types of objects, including: multimedia content publishers and target objects associated with the multimedia content publishers. The parameters of the first association processing module are updated based on the association relationships between various types of training objects, including: The association between training multimedia content includes an initial association and an additional association obtained by a second association processing module based on the features of the training multimedia content. The initial association is determined based on the training multimedia content.

15. A device, characterized in that, The device includes a processor and a memory; The processor is configured to execute instructions stored in the memory to cause the device to perform the method as described in any one of claims 1 to 13.

16. A computer-readable storage medium, characterized in that, Includes instructions that instruct the device to perform the method as described in any one of claims 1 to 13.

Images